Table 1.
Shows the location details regarding the data gathering in Delhi.
Fig 1.
The general architecture of the proposed model of regression.
Fig 2.
Density plot of PM2.5 for multi station.
Fig 3.
The mean normalized air quality data across multiple stations.
Fig 4.
A Spearman’s ranking correlation coefficient represents the Mean correlation between PM2.5 and atmospheric components, for multi-stations.
Table 2.
Spearman’s rank means correlation coefficient between PM2.5 and Climatic Parameter across multiple stations.
Fig 5.
Comparison of original and wavelet-transformed features across six monitoring stations.
Fig 6.
Principal component analysis (PCA) of air quality data across six monitoring stations.
Fig 7.
Radar chart of feature importance.
The significance of seven factors for forecasting PM₂.₅ at six Delhi stations. Lines further from the center imply more importance.
Table 3.
The section that follows the pseudo-code for the hybridized AOAOA technique.
Fig 8.
The sequence diagram illustrates the implementation of the hybrid AOAOA approach.
Fig 9.
Generic network architecture of the LSTM.
Fig 10.
Air quality forecast is based on the Bi-LSTM framework.
Table 4.
Model’s architecture, compilation, training, and evaluation metrics.
Fig 11.
Graphical representation of MSE measures of several epochs of AquaWave –BiLSTM.
Fig 12.
Graphical representation: actual vs. predicted PM2.5 for multiple stations.
Fig 13.
Visual representation of R² score and MSE comparison across stations with and without feature extraction.
Fig 14.
Multi-station performance comparison of feature selection methods for air quality prediction: AOAOA vs. competing algorithms.
Table 5.
Analysis of several machines and deep learning algorithms for multi-station air quality datasets.
Table 6.
Top-ranked SHAP feature (Rank 1) at each station, its PCA component, and the three most significant meteorological and pollutant factors influencing PM2.5 prediction. PM₁₀ emerged as the primary predictor across stations, with RH, AT, AP, and WS also making substantial contributions.
Fig 15.
SHAP summary plots of the top 10 features contributing to PM2.5 prediction at each station: (a) AshokVihar, (b) DCStadium, (c) DwarkaSec8, (d) Najafgarh, (e) NehruNagar, and (f)Okhla.
Table 7.
Station-wise MSE comparison across proposed and baseline methods.
Fig 16.
Comparison of average mean squared error (MSE) across feature extraction and selection methods for multiple stations.